The Numbers Worth Watching
Use digital records to locate a process problem, not to make an unsupported judgment about a person. That distinction should be written into any evaluation of visit the provider and reflected in who can access the resulting data.
When this issue touches live workplace policy, compare the local evidence with guidance published by CIPD before changing a process or communicating a rule.
Seven figures that describe whether the arrangement works, and the popular ones that mostly describe the labour market.
Recruitment and retention generate a lot of reportable material, most of which describes conditions rather than management.
The seven
Ninety-day survival, by source and by cohort.
Attrition by week of tenure, as a distribution rather than an annual percentage.
Cost per person present at ninety days.
Attrition by team, which is the supervisor cut.
Time from application to first shift.
Proportion of starters reaching standard within the planned ramp.
And returner rate, where the operation has a peak.
What each says when it moves
Ninety-day survival falling with stable attrition overall: the intake has changed, usually the source or the advert.
Week-one attrition rising: induction, equipment, or the first-day logistics.
Week-three attrition rising: the ramp, the first payday, or the training rate nobody disclosed.
Attrition concentrated in one team: a supervisor conversation, avoided for years in most operations.
Time to first shift lengthening: people are taking other jobs while they wait, which is why the campaign is under-delivering.
The figures that mislead
Annual attrition percentage, which mixes first-week leavers with five-year leavers and describes neither.
Cost per hire, which rewards filling seats with whoever is available.
Time to fill, the same.
Headcount against establishment, which says nothing about whether the people are trained.
And engagement survey scores from a workforce that turns over three times a year, which measure the survivors.
Cuts that are worth the effort
Direct against agency, always separated.
Peak hires against permanent.
By shift pattern.
By source.
By supervisor.
Five cuts, all available from data you hold, and each answers a question the headline number cannot.
Producing them without a system
All seven come from the payroll record and the start list.
Start date, leave date, source, shift, team.
Five fields, a spreadsheet, and an afternoon.
An operation waiting for a workforce analytics platform measures nothing for a year, and the platform will produce the same seven numbers from the same five fields.
Who gets what
The operations manager: attrition by team and by tenure week.
Recruitment: ninety-day survival by source, and cost per ninety-day survivor.
Supervisors: their own team's figures, as information rather than as a performance measure.
And the workforce: what the exit data said and what changed, which is the part that makes the next round honest.
The annual question
Did ninety-day survival improve?
Did the first-fortnight spike shrink?
Was anything done about the team with the worst figure?
And would you recruit the same way again?
Four questions, and a report that cannot answer them has described the labour market rather than the operation.
Setting a baseline
Before changing anything, record the seven as they are.
Because without it nothing afterwards can be attributed, and every improvement will be credited to whatever was most visible.
An afternoon, at the point where a delay feels least affordable.
The one to publish
Ninety-day survival.
To recruitment, to operations, and to the supervisors.
Because it is the only figure that aligns them: recruitment stops being rewarded for volume, operations sees the cost of the induction they skipped, and supervisors see their own team.
A single number, shared, changes more behaviour than any policy document.
Reporting upward
One page: ninety-day survival, the tenure distribution, cost per survivor, and what changed.
Plus one honest statement of what has not improved, which is what makes the rest credible and is the section most often cut.
Producing them from five fields
Start date, leave date, source, shift, team.
That is the whole requirement. An operation waiting for a workforce analytics system measures nothing for a year, and the system will produce the same seven numbers from the same five fields.
How it fails in practice
The figures are produced monthly, circulated, and read by nobody, because none of them is attached to a decision anybody has to make..
The check
When a number last moved, did anything change? If the answer is no for two quarters, the problem is ownership rather than measurement..